Anthropic is investigating reports that a small group of unauthorized users gained access to its Mythos AI model. The technology has not been released to the public because the company considers it too sensitive to release, warning that its ability to detect cybersecurity vulnerabilities could enable dangerous cyberattacks.
The unauthorized users reportedly accessed the Claude Mythos Preview by guessing the model's URL on the same day Anthropic announced Project Glasswing. Communicating through a private Discord channel, the group has since used the model regularly, though not for cybersecurity purposes.
Anthropic confirmed it is investigating the claims but maintains that there is no evidence its core systems have been impacted. The incident highlights the risks associated with restricting access to frontier AI capabilities.